Prediction accuracy against the predictor's Elo advantage

One point per ordered pair of models; point area grows with the number of scorable predictions. The line is a prediction-weighted logistic fit with a 95% bootstrap band.

Odds of a correct prediction change by ×0.94 per 100 Elo of advantage (95% CI 0.87–1.01). Stronger-than-opponent predictors were right 55% of the time; weaker ones 62%.